<!doctype html public "-//IETF//DTD HTML//EN">
<HTML>

<HEAD>

<META NAME="GENERATOR" CONTENT="Internet Assistant for Word 1.0Z">
<META NAME="AUTHOR" CONTENT="C. Hooton">
</HEAD>

<BODY>

<H1>INTRANATIONAL POLITICAL INTERACTIONS (IPI)</H1>

<H1>CODING RULES</H1>

<H4>Principal Investigators:</H4>

<P>
David R. Davis, Department of Political Science, Emory University
<P>
&amp; Will H. Moore, Department of Political Science, University
of California, Riverside
<H4>Coding Rules Authors:</H4>

<P>
Brett Ashley Leeds, David R. Davis, and Will H. Moore, with Christopher
McHorney 
<P>
20 September 1995 
<P>
Version 7.3 <HR>

<P>
TABLE OF CONTENTS
<P>
1. Introduction 
<P>
1.1 The Goals of the Project p. 4 
<P>
1.2 Comparison with Other Event Datasets p. 4 
<P>
1.3 Conceptualizing Conflict p. 6 
<P>
1.4 Conceptualizing Cooperation p. 6 
<P>
2. Coding Events 
<P>
2.1 Defining a Political Event p. 7 
<P>
2.2 Actor and Target Assignment p. 8 
<P>
2.3 Multiple Actors and Targets p. 10 
<P>
2.4 Coding Multiple Events p. 10 
<P>
2.5 Dating the Events p. 13 
<P>
2.6 Controversial Reports p. 14 
<P>
2.7 Distinguishing Domestic Events from International Events p.
14 
<P>
2.8 Sources for Events Data p. 15 
<P>
2.9 Ambiguous Values p. 15 
<P>
2.10 The 'Deaths' Variable p. 15 
<P>
2.11 The Coding Sheet p. 17 
<P>
2.12 Managing Computer Files p. 19 
<P>
3. Actor and Target Codes 
<P>
3.1 Selecting Actor and Target Codes p. 20 
<P>
3.2 The State p. 21 
<P>
3.3 Mass/Elite Distinction p. 22
<P>
3.4 Unspecified Actors and Targets p. 22 
<P>
3.5 Business and Labor p. 23 
<P>
3.6 Multinational Groups and Foreign Nationals p. 23 
<P>
3.7 Leftists and Rightists p. 24 
<P>
3.8 Religious Groups p. 24 
<P>
3.9 Socially Influential Individuals p. 25 
<P>
3.10 Supranational Organizations p. 25 
<P>
3.11 Case Specific Codes p. 25 
<P>
3.12 Distinguishing Among Ethnic Groups, Social Groups, and Dissident
Groups p. 26 
<P>
3.13 Coding Factions and Coalitions p. 26
<P>
4. The Conflict Scale 
<P>
4.1 Introduction to the Conflict Scale p. 27 
<P>
4.2 Press Criticism of the Government p. 27 
<P>
4.3 Rumors of Threatening Action p. 27 
<P>
4.4 Strikes, Protests, and Riots p. 28 
<P>
4.5 Executive Adjustment, Political Resignations p. 29 
<P>
4.6 Censorship p. 29 
<P>
4.7 Mobilizing Resources p. 30 
<P>
4.8 Violent Political Acts p. 30 
<P>
4.9 Mass Arrests and Political Convictions p. 30 
<P>
4.10 Bombings p. 31 
<P>
4.11 Assassinations p. 32 
<P>
4.12 Reports of Disappearances p. 32 
<P>
4.13 Breaking a Truce, Breaking a Peace Treaty p. 32 
<P>
4.14 Attacks, Clashes, and Battles p. 32 
<P>
4.15 Occupation of Territory p. 33 
<P>
4.16 Coups d'Etat p. 33 
<P>
5. The Cooperation Scale 
<P>
5.1 Introduction to the Cooperation Scale p. 34 
<P>
5.2 Promises p. 34
<P>
5.3 Relaxation of Repression p. 35 
<P>
5.4 Political Reform p. 35 
<P>
5.5 Talks, Agreements to Talk p. 36 
<P>
5.6 Surrenders, Releases of Prisoners or Hostages p. 36 
<P>
5.7 Cessation of hostilities p. 37 
<P>
5.8 Elections p. 37 
<P>
6. Using the Conflict and Cooperation Scales Together 
<P>
6.1 Using the Scales Together p. 38<HR>

<P>
INTRODUCTION
<P>
1.1 The Goals of the Project
<P>
The Intranational Political Interactions (IPI) project is designed
to measure political conflict and cooperation within societies
through the coding of political event reports from international,
regional, and local sources. These events are coded on two ten
point scales which reflect the severity of various cooperative
and conflictual statements and actions. We use this scaled events
data to calculate the volume and intensity of political conflict
and cooperation within the domestic polity. In addition to facilitating
the calculation of general levels of political conflict, the IPI
coding scheme allows the examination of the dynamics of interaction
among specific groups within the society. While scholars have
focused a great deal of attention on the causes of political violence,
internal war, and revolutions, their explanations have typically
been based on attributes of the parties involved and attributes
of the socieconopolitical environment in which these conflicts
occur. As a consequence, the literature has failed to address
the impact of the behavior of these groups and, more importantly,
the interactions between groups. IPI gives scholars the ability
to track interactions among social groups and between the state
and social groups.
<P>
1.2 Comparison with Other Events Datasets
<P>
Two data collection efforts begun in the 1960s [COPDAB (Azar,
1982) and WEIS (McClelland, 1972)] have given scholars of international
relations a useful bank of information with which to examine issues
of conflict and cooperation among nation states in the international
system. Referred to as &quot;events data&quot;, these data record
the behavior of nations interacting with one another. Burgess
and Lawton define events data as &quot;words and deed--i.e. verbal
and physical actions and reactions--that international actors
(such as statesmen, national elites, intergovernmental organizations
(IGOs), and nongovernmental international organizations (NGOs)
direct toward their domestic or external environments&quot; (1972,
p. 6). The availability of this data has spawned a large literature
examining peace and conflict in the international system. Unfortunately,
similar data have not been available for the study of conflict
and cooperation within nation states. The IPI project is an attempt
to fill that void.The only existing dataset which attempts to
measure the behavior of domestic actors is the &quot;Daily Political
Events Data&quot; from The World Handbook of Political and Social
Indicators (Taylor &amp; Jodice, 1983). The World Handbook provides
information on several classifications of conflict relevant events,
including four types of demonstrations, six types of armed attacks,
two types of strikes, seven types of leadership changes, five
types of elections, nine types of state action with respect to
civil liberties, and finally the number of riots and the number
of deaths from domestic political violence. While this is an impressive
collection of variables, it is only a typology. No scaling system
for the events analogous to those used in the international events
data is developed, and, more importantly, only a small number
of types of events are coded, thus limiting the utility of the
data for the comprehensive study of political conflict in a society.
The World Handbook scheme ignores, for example, statements, speeches,
and petitions. Clearly, these events are relevant to the study
of internal interactions (i.e., the opposition makes demands against
the state, the state denounces the opposition, etc.). The World
Handbook also fails to track cooperative interactions between
groups embroiled in conflict. If we are interested in modeling
conflict processes and testing those models, we need data that
account for the full spectrum of relevant events.In addition to
the type of event, the Handbook provides information on the actor,
the target, and the issue that motivated the event. Taylor and
Jodice clearly intended to make their data useful for testing
hypotheses regarding the behavior of actors and their interactions,
but their choice of actor and target designations leads one to
believe that they were envisioning crossnational studies rather
than timeseries case studies. For example, there are eight &quot;actor
groups&quot; variables: government, political party, political
group, military, clergy, intellectuals, workers, students, minorities,
revolutionaries, and general population. While these categories
are useful for broad comparisons across many cases, they are less
useful for the analysis of specific countries. In most states,
for instance, there are several political parties pursuing very
different policies, but the Handbook coding scheme fails to distinguish
among them.The IPI project is designed to overcome some of the
limitations of the World Handbook scheme. First, IPI data are
scaled according to intensity. Second, IPI provides codes for
a broader range of political events, both verbal and substantive,
both conflictual and cooperative. Third, IPI assigns unique codes
for actors and targets which may be aggregated into broad categories
for cross-national studies, but also can be used for detailed
time series analysis of the interactions among particular groups.
Fourth, IPI data serves to update and increase the temporal domain
of the World Handbook data. Many of the World Handbook variables
cover only the period 1968-1982. Because the IPI coding scheme
remains compatible with World Handbook data, the two datasets
together provide a consistent longer term time series.We note
that the Handbook is not the only source for data on internal
conflicts. Those familiar with the COPDAB project (Azar, 1982)
recall that it includes data on conflict and cooperation within
nation states. Unfortunately, the utility of the domestic COPDAB
data is limited by the fact that it is aggregated at the national
level. As a result, the data indicate an overall level of cooperation
and conflict in a given nation state, allow no indication of the
extent to which the state or opposition groups are instigating
the violence. It is the ability to distinguish among actors that
drives the IPI project.The IPI coding scheme represents our attempt
to mirror the successful data collection efforts at the international
level using domestic actors as the unit of analysis. In the process,
we have overcome most of the limitations of existing domestic
political event datasets by 1) increasing the number of actor/target
codes to allow users flexibility in choosing appropriate levels
of aggregation, and 2) coding events with respect to intensity
on ten point scales for cooperative and conflictual interactions.

<P>
1.3 Conceptualizing Conflict
<P>
Conflict has two crucial elements. First, conflict involves perceptions
of incompatible interests. Groups in conflict believe that the
realization of one party's interests inhibits the realization
of the other party's interests. Second, conflict is reflected
in actual behavior; we are not interested in &quot;latent&quot;
or &quot;potential&quot; conflict. One or both parties take purposive
action against other parties in an effort either to advance their
own interests or inhibit the realization of the interests of others.
In other words, conflict is visible in actions and events. As
Tilly states, conflict tends to flow directly out of a population's
central political processes, instead of expressing diffuse strains
and discontents within the population; . . . the specific claims
and counterclaims being made on the existing government by various
mobilized groups are more important than the general satisfaction
or discontent of those groups, and . . . claims for established
places within the structure of power are crucial (Tilly, 1985,
p. 1).Thus, the actions undertaken by groups to advance their
own interests at the expense of others or to inhibit the realization
of the interests of another group are a valid measure of the level
of political conflict within a society.
<P>
1.4 Conceptualizing Cooperation
<P>
Cooperation has historically been thought of as the absence of
conflict. Our concept goes beyond this simple definition. We believe
that cooperation, like conflict, involves intentional, purposeful
action in pursuit of goals. Keohane's distinction between cooperation
and harmony is especially useful. Harmony refers to a situation
in which actors' policies (pursued in their own self-interest
without regard for others) automatically facilitate the attainment
of others' goals. Cooperation, on the other hand, requires that
the actions of separate individuals or organizations-- which are
not in preexistent harmony-- be brought into conformity with one
another through a process of negotiation which is often referred
to as 'policy coordination' . . . . Cooperation occurs when actors
adjust their behavior to the actual or anticipated preferences
of others, through a process of policy coordination (Keohane,
1984, p. 51). Groups pursue their goals through both conflictual
and cooperative actions. Understanding the dynamics of political
interactions thus requires measuring both the conflictual and
cooperative behavior of domestic actors. The identification of
actors and the categorization of their behaviors lies at the heart
of the IPI coding scheme. 
<P>
2. CODING EVENTS
<P>
2.1 Defining a Political Event
<P>
With respect to the conceptual definition of an event, this project
adopts a modified version of the Azar (1975) definition. We are
interested in cooperative and conflictual political events: A
political event is an action taken by an actor at a given point
in time to advance its political interests. Thus a political event
involves (1) an actor, (2) a target, (3) a time period, (4) an
action, and (5) a political interaction. By political, we mean
issues that involve the authority to make decisions concerning
the extraction and distribution of social resources or values.
We replace Azar's concern about issues with the requirement that
these events involve a political interaction. Operationally, we
modify Gerner et al. (1992) and define a political event report
as: A natural human language statement describing a political
event, where the (sometimes implied) subject and object(s)/direct
object(s) denote an actor/target and the verb denotes an action
taken by the actor.Azar additionally specifies that events must
be &quot;distinct enough from the constant flow of 'transactions'
. . . to stand out against this background as 'reportable' or
'newsworthy'&quot; (Azar, 1982, p. 3). Howell interprets this
to mean that &quot;Transactions are the normal activities between
nation-states, while interactions are the extraordinary events
that occur in irregular patterns&quot; (1983, p. 150). We, too,
are interested in extraordinary events (interactions), rather
than normal activities. Thus, opinion polls, the imposition of
austerity measures, standard labor-business negotiations, and
other similar actions which represent the normal functioning of
government or society are not codable under the IPI scheme. Reactions
to any of these events, however, may be codable. Protests in response
to austerity measures, for instance, are coded.In some instances,
the line between normal activity and an extraordinary event is
fine. Democratic elections are important and unusual events in
some societies, but in established democracies they represent
no more than the normal functioning of government. We code all
activities surrounding the first two elections held in any state
undergoing a transition to democracy with the exception of personal
attacks against candidates. Grandstanding and content free statements
(e.g., criticisms of an official's family life, nonspecific campaign
promises) are never codable in the IPI scheme. In established
democracies, we do not code campaign rallies for individual candidates
unless they become explicitly pro- or anti-government demonstrations,
and we do not code the elections themselves. We do, however, continue
to code all statements regarding policy. Similarly, it is sometimes
difficult to isolate the circumstances under which conflict between
governmental branches is extraordinary and should be coded and
when disputes represent the normal policy making process. We code
all conflict between governmental branches which relates to institutional
questions. In other words, when conflict between the legislature
and the executive relates to the definition of areas of jurisdiction
and responsibility, we code the conflict. When the legislature
and the executive dispute policy, however (for instance, the legislature
passes laws which the executive does not favor), we do not code
the activity. Policy statements by political actors, however,
are codable. If the leader of a political party criticizes the
executive's anti-inflation plan and vows to defeat it in Congress,
we code a conflictual statement from that party to the executive.

<P>
2.2 Actor and Target Assignment
<P>
The term actor carries with it a connotation of initiator. We
wish to note explicitly that we do not intend that connotation
in our use of the term. It is frequently impossible to ascertain
who started a set of given interactions, and no attempt is made
to do so. Rather, we begin by assuming that groups face options
to act or not to act at any given point in time. If they act,
they are coded as actors. If they are acted upon (i.e., the recipient
or target of another party's action), they are coded as targets.
We have created (and attached) a list of generic actor-target
codes to facilitate defining actors and targets. For each country
we examine, the generic sheets are modified to reflect the specific
circumstances and political context of that case (see section
3.11). 
<P>
Consider some examples from The New York Times Index's reporting
of events in Colombia, 1965: Bandits attack peasant farms, Paime,
Cundiamarca Dept., kill 5, wound others, Je 2, 5:5. Here, the
actor is bandits and the target is peasants. Since there is no
account of the peasant farmers responding to the attack with violence,
they are not coded as actors with the bandits as targets. If such
a report were noted, however, each party would be coded as an
actor and as a target. That is, when two parties interact, both
are assumed to have the option of acting: in this case, the peasants
can respond to the attack, or they can run and take cover. Since
the report does not explain that they attacked, we assume that
they took cover. However, had the words &quot;peasant farms&quot;
been &quot;police station&quot;, the same judgment would not have
been made. Different parties to a conflict have different 'base
line' responses. We adopt the simplifying rule that if a party
to a conflict event is known to be armed, then we assume that,
unless reported otherwise, the party will respond to an attack
or ambush by firing upon its assailants. Since peasants are not
generally known to be armed in Colombia, their `base line' response
is assumed to be flight. However, guerrillas, bandits, police,
and the armed forces are assumed to have a 'base line' response
of attack. 
<P>
Thus, the following report is coded as explained below: Police
kill bandit chief J B Ruiz Castro, Jl 7, 14:3. Here, the police
and the bandits are coded as both actor and target (i.e., two
events are coded whereas in the attack on the peasant farms only
one event was coded). Since a bandit surrender would be unusual
(and would likely not lead to the chief's death, but rather to
his capture), we invoke the assumption that the reporters would
have noted it had the bandits surrendered. Certainly, this assumption
will be poor in some cases. However, we contend that failure to
invoke this assumption would lead to one-sided coding of two-sided
events much more frequently than the assumption leads to two-sided
coding of one-sided events (i.e., invoking the assumption reduces
the noise in the data).
<P>
The first report above raises another question about the actors
and the targets. We are interested in political conflict events.
Are 'bandits' political actors? Unfortunately, this is a thorny
issue because government actors who provide reporters with the
information they use to file their reports have a significant
interest in minimizing the political claims of their opponents.
Thus, they sometimes refer to political opponents who have taken
up arms as bandits. As coders, we face three choices: 1) we can
code them as guerrillas, 2) we can ignore bandits as actors, 3)
we can code them as bandits and allow users of the dataset to
determine whether they want to include bandits in their study.
We find the third option vastly superior to the alternatives.Sometimes,
coding the target can represent a problem, as in the following
example: Govt reports 11 dead in guerrilla attack; repts 14 others
dead in Dec 31 attacks. From this report it is unclear who is
the target and, as a consequence, how many events should be coded.
Unfortunately, this record is incomplete. We know that at least
two events took place, that the actor in both cases was unspecified
guerrillas, and that eleven and fourteen people died in either
armed attacks or military clashes, depending on who the target
was. Since it is better to err on the conservative side, we code
it as an armed attack (604 on the conflict scale) on an unknown
target (99) and indicate in our event record that we need more
information. We hope that as we code additional sources, more
facts will be reported which will allow us to fill in the missing
information, although there will be a substantial number of events
for which this is not the case.It is important to note that victims
are not always targets. This distinction becomes important in
coding terrorist or repressive events. The state assumes responsibility
for maintaining public order. Terrorist events like bombings and
hijackings challenge the ability of the state to fulfill this
responsibility. While bystanders may be injured or killed and
suffer as victims of such events, in most cases, we consider the
target of the event to be the state. Thus, if a foreign embassy,
or an electrical pylon, or a hotel is bombed, the target is the
state. If stores are robbed during a riot, the target of the event
remains the state, not business elites. The exception to this
rule is terrorist events which are clearly aimed at disadvantaging
other groups, for instance a bombing of a dissident group's headquarters,
or of the home of a trade union leader, or of a group of businesses
owned by members of a particular ethnic or religious group. In
these cases, the target of the event is the disadvantaged group.

<P>
The argument can be extended to repressive events. In many cases,
the individual victim of repression is not the target of state
repression. Rather, repression is used as a tool for intimidating
the entire society rather than the individual. If hundreds are
disappeared by the state, therefore, we code the target as 'general
population' (09), regardless of the affiliations of the victims.
Once again, however, if repressive actions are aimed at disadvantaging
particular groups, those groups are considered targets. Thus,
the murder by death squads of a single trade union leader is an
action targeting labor unions. Occasionally, an action will be
aimed at a social group, but will additionally challenge the ability
of the state to maintain order. In this case, the event is coded
with multiple targets. Consider the following paraphrased report
from Reuters News Wire, November 21, 1986, on events in Mexico:
During the parades, supporters of the PRI violently clashed with
members of the PAN. Seventy were wounded as police used teargas
to break up the rioting. In this case, the PRI is acting violently
toward the PAN and the PAN is acting violently toward the PRI.
Both the PAN and the PRI are also challenging state order. Thus,
this report would result in six event records: two records representing
the violence from each of the parties to the other (401 on the
conflict scale), two representing the violence to the state from
each of the parties (401 on the conflict scale), and two representing
state violence to the parties (402 on the conflict scale). 
<P>
2.3 Multiple Actors and Targets
<P>
There are a number of instances in which multiple actors or targets
are identified in a given report. Consider the following fictitious
example: In a show of solidarity, government employees, teachers,
laborers and peasants marched on the capital today to protest
recent government cut-backs in wages and prices.Here we have a
single target (the state--10) but several actors: government employees
(53), teachers (48), laborers (56) and peasants (50). Thus, the
coder codes four conflictual events: a protest (301 on the conflict
scale) for each of the four social actors against the state. In
addition, the coder accounts for the cooperation between the four
groups acting together. In this case, each of these four social
groups is cooperating with other social groups, so the coder codes
four 304's on the cooperation scale, one for each social group,
with the target of the cooperative action listed as 'unspecified
social actors' (75). 2.4 Coding Multiple EventsOften multiple
events appear in a single news report. Coders need to pay strict
attention to this issue, as it accounted for the vast majority
of the unmatched coding in the first round of coding for Moore's
&quot;Violent Intranational Conflict Data Project: Phase I&quot;
(Moore and Lindstrom, 1994). 
<P>
One example was covered in the second event noted above (section
2.2): a report of one group attacking another should be coded
as two events if the group that is attacked is generally known
to be armed, unless it is expressly reported that the other group
did not respond in kind. A second example appears in section 2.3.
When several groups act together, we code a separate event for
each actor/target combination. A third issue concerns references
to past events in a report that is primarily concerned with a
present event. A previous report (section 2.2) provides an obvious
example; it is noted that there were &quot;25 others dead in Dec
31 attacks&quot;. The coder must be sure to enter this event as
well. 
<P>
A less obvious example is this report from Colombia, 1956 (TNYTI):
dispatch from Quito repts 80-200 villagers slain by Army troops
in May near Chaparal, Tolima Prov, in reprisal for death of a
few soldiers ambushed by guerrillas;.... Je 12,1:5. In this example
there are three events: 1) the army's slaying of peasants in May,
2) guerrillas ambushing the army at an unspecified date, and 3)
the army attacking guerrillas on that same unspecified date. We
discuss problems associated with dating events below (section
2.5).Occasionally, journalists offer summary reports of events
which have occurred during a recent time period. For instance,
a report might state, &quot;fifty bombings have occurred in the
last month in Chile&quot;. This report is coded as fifty separate
events, albeit with much missing information. The coder checks
the 'Needs More Information' field for each of these fifty event
reports with the hope that more information can be located in
other sources.A fourth issue concerns events that endure for more
than twenty-four hours. General strikes, riots, etc. often take
place for several days and even months. In this project we note
that all human beings live on a twenty-four hour schedule because
our bodies simply require that we get some sleep, nourish ourselves,
and attend to our bodily functions. Thus, no human beings sustain
strike activity, rioting, or other events for days on end without
a break. Since the rising and setting of the sun provide a convenient
marker, we set the rule that no single event can last more than
twenty-four hours. Thus a twelve day strike is coded as twelve
separate events. 
<P>
As an example, consider the following paraphrased report from
Lebanon, 1974 (TNYTI): Police storm campus, kill 2, terminate
12 day strike. It was the first mention of the student strike,
and hence the coder must code twelve strike events by the students
against the state (302, conflict scale). In addition, the coder
must code one 'Police violence in response to demonstration with
deaths' event (504, conflict scale) with students as the target.
Finally, the coder will note two deaths from political violence
(see section 2.10).Any time a long term political standoff results
in daily interactions, an event is coded for each day of its duration.
Thus, if a group occupies a building and holds hostages for five
days while negotiating with the government, we code a violent
political act (505 on the conflict scale) for each of the five
days. Similarly, battles, strikes, protests, and riots are coded
for each day of their duration. There are other situations which
can not be assumed to result in daily interactions, however, and
these events are coded only as they begin, end, and experience
reported changes. 
<P>
Examples include most repressive actions (for instance, the imposition
of states of emergency, the closing of universities, and the imposition
of censorship), kidnappings, and occupation of territory by the
opposition. A difficult situation arises, however, from the following
type of report (Philippines, 1975, TNYTI), which is typical of
statements where body counts of deaths from violent political
conflict are announced by state officials: announcement coincides
with new fighting with 15 Govt soldiers reptd killed in last 10
days with guerrillas in provinces of Sulu, Cotabalto and Zamboanga;
at least 5 rebels are known dead and mil announced that some 500
civilians were rescued from Zamboanga del Sur where they had been
held hostage; map (M), Ja 11, 3:1. 
<P>
We know that more than one clash occurred between the state and
guerrillas during this ten day period, but we do not know how
many. In the absence of better information, we assume that fighting
occurred on half of the ten days in each of the three locations
(total thirty military clashes, fifteen with the state as actor
and fifteen with the guerrillas as actor). The coder codes military
clashes for five days and checks 'Needs More Information' on the
coding sheet in the hope that information from another source
might clarify the report. The twenty-four hour rule is useful
because it enables us to build duration into the coding without
creating a separate scale (e.g., Taylor and Jodice use a scale).
Some might argue that this scheme is biased toward long events,
that because it counts each day as a new event, events with long
duration overwhelm other events of shorter duration. This argument
is predicated on the idea that there is a decreasing marginal
impact by day of an event as it persists over time. 
<P>
Our coding scheme does not preclude end users from taking such
an argument into account. Anyone using the data may transform
the data on an event-by-event basis by identifying those events
that occurred over multiple days, taking the natural logarithm
of all events that persist beyond some number of days, and using
that log-transformed figure in his or her analysis. This scheme
enables the end-user to make those decisions. Finally, political
conflict events have a spatial as well as a temporal dimension.
Just as we use twenty-four hours as the natural temporal unit,
we use the village/town/city as the natural spatial dimension.
If the same event occurs simultaneously in one village, four towns
and two cities, then we code seven events (for that day). To summarize,
events are identified with regard to actors and targets, time,
and space. If two groups protest in three cities for five days,
a total of thirty (two times three times five) events are coded.Occasionally,
however, a report will seem to indicate that several codable events
have occurred when really one encompasses the others. 
<P>
Consider this example from the Latin America Weekly Report, March
26, 1982: At least two soldiers and six M-19 guerrillas were killed
in a series of clashes in the southern departments of Cacqueta,
Huila and Cauca last week. Guerrillas blew up a bridge and high-tension
pylons on the Florencia-Neiva road, and ambushed military vehicles.
Fighting was continuing this week, as troops of the Juanambu battalion
moved in (p. 12).In this case, we code twenty-four military clashes
(704) because we have reports of two actors fighting in three
locations over four days (We invoke the assumption that fighting
occurred on half of the days mentioned in the report--last week
continuing into this week = eight). We do not, however, code bombings
and ambushes as well. The military clash encompasses these activities.
Similarly, if a group protests in front of the seat of government
shouting anti-government slogans, we code the protest (301), but
we do not code additional verbal conflict events (100 and 200
level) (see section 4.4).
<P>
2.5 Dating the Events
<P>
The events are assigned the date on which they are reported to
have occurred, not the day on which they are reported. In some
cases, a precise date is impossible to assign. When this happens,
we record as much information about the date as we can. For instance,
if we do not know exactly which day the event occurred, but we
do know which week, we record the week. In this case, the entry
in the &quot;Event Date&quot; field appears in the following form:
JAN w1 1989. We follow the GEDS (Davies, 1993) rules for assigning
weeks. They are as follows: week 1: 1-7 week 2: 8-14 week 3: 15-22
week 4: 23-end.
<P>
When we cannot even determine the week of the event, we simply
assign the missing value score '99' for the day variable and then
seek to obtain a precise date from a different source. (The coder
should check 'Needs More Information' on the coding sheet.) In
most cases, the month in which the event occurred will be apparent.
If the month can not be reliably isolated, then we record the
quarter using the following format: Q1 99 1989. quarter 1: January-March
quarter 2: April-June quarter 3: July-September quarter 4: October-December
<P>
While missing data on the 'day' variable do not present a problem
for people who aggregate the data at monthly, quarterly, or annual
units of time, we wish to avoid using the missing data value '99'
in the 'month' variable whenever possible.The &quot;dispatch from
Quito&quot; example above (section 2.4) highlights a dating problem
where it is necessary to invoke some assumptions. We assume that
parties respond quickly to wrong-doing so that, given no other
information, when a report notes that a party is responding to
a previous event, we will assume that event occurred within the
past five days. Thus, if the reported event occurred on or after
the fifth day of the month, the coder dates the event to which
it is a response as occurring during the same month, with the
day coded as missing (i.e., 99). If the reported event occurred
prior to the 5th day of the month (i.e., 1st-4th), the coder dates
the event to which it is a response as occurring during the previous
month, again with the day coded as missing (i.e., 99). In these
cases, the 'Needs More Information' box is checked in the hope
that the missing date will be clarified by secondary sources.

<P>
2.6 Controversial Reports
<P>
Sometimes it is difficult to tell whether an event actually happened.
Consider the following report from Latin America Weekly Report,
February 26, 1982: The government claims that six more members
of M-19 gave themselves up to the southern command of the army,
in Florencia, Cacqueta, on 22 February, three days after the publication
of the latest official amnesty offer (p. 11). By using the word
&quot;claims&quot; the source suggests some doubt as to whether
these surrenders actually occurred. We elect to code all such
reports as actual events except where the source explicitly notes
controversy over the event's occurrence. We code this report as
a surrender (see section 5.6 for further information about coding
surrenders). Had the report continued with the statement, &quot;Leaders
of the M-19 deny that these surrenders occurred and continue to
reject the amnesty offer&quot;, however, we would not code a surrender.
Instead we would code a verbal conflict event (101) from the state
to the M-19, and another from the M-19 to the state. 
<P>
2.7 Distinguishing Domestic Events from International Events 
<P>
Some events which may initially appear to be international events
are codable as domestic political conflict or cooperation in the
IPI coding scheme. One example was given above in section 2.2.
When a foreign embassy is bombed, the event can be coded as domestic
political conflict between the actor and the state, which is responsible
for maintaining order in society. In addition, when domestic groups
protest against the influence of a foreign power with whom the
state is allied or against alliance with a foreign power, the
event can be recorded as conflict between the domestic group and
the national government. 
<P>
Two examples of such activities are given in the GEDS (Global
Events Data System) Coders' Manual (Davies, 1993): A militant
Islamic group demonstrated in Cairo for an end to the alliance
with US forces in Saudi Arabia (p. 27). Hundreds of Filipino students
demonstrate outside the US Embassy in Manila for the withdrawal
of American bases (p. 28).In both of these instances, the events
can be coded as protests (301, conflict scale) with a social group
as actor and the state (10) as target. Similarly, when members
of a polity storm a foreign embassy requesting asylum, the event
can be coded as a protest (301 on the conflict scale) against
the national government. Events which happen outside the borders
of the state in question are occasionally codable in the IPI system
if they are explicitly related to a political conflict within
the state. When a dissident organization is working to unseat
the government of a state from the territory of a neighboring
state, we do code interactions between the dissident group and
the government it is opposing. 
<P>
For example, if it is reported that Colombian drug lords have
assassinated the Colombian ambassador to Hungary in Budapest,
the assassination is codable as a domestic political event in
Colombia. When a Pakistani dissident group hijacks a plane in
Pakistan and flies to Syria where a ten day standoff with the
Pakistani government ensues, we code domestic political conflict
for each day of the standoff. When the Indonesian government raids
an Indonesian dissident group's settlement in Papua New Guinea
in response to the group's attacks in Indonesia, this too is codable
under the IPI system.There are also cases, however, when political
conflict events which occur in the territory of the state being
coded are not domestic political events relevant for that state.
For example, if Afghan students studying in India protest in front
of the Afghan embassy in India to show their objection to the
government of Afghanistan, this is not codable as an IPI event
for India. It is codable as an IPI event for Afghanistan.
<P>
2.8 Sources for Events Data
<P>
In order to improve the density and quality of our data collection
effort, numerous local, regional and international news sources
will be coded during the project. In the early stages of the project
we are concentrating on Reuters News Wire. As the scope of the
project expands, we anticipate adding additional local and regional
sources. Current source codes are as follows: 1. The New York
Times Index 2. Reuters News Wire 3. Latin America Weekly Report
<P>
2.9 Ambiguous Values
<P>
Consider the following fictitious report: State authorities launched
a nationwide sweep yesterday, arresting 90 members of a well known
dissident group. The operation required police to coordinate their
efforts in the cities of X, Y, and Z as well as several municipalities.We
know in this instance that arrests should be coded for three cities
and for &quot;several&quot; municipalities. News reports are full
of terms like 'several', 'scores', 'hundreds', and other vague
terms. In this project we arbitrarily (but consistently) assign
the following values to these terms: couple 2 few; some 3 several
5 many; numerous 15 dozens 36 scores 60 hundreds 250 thousands
2500. Thus, in the above example, eight events are coded because
we have assigned the word 'several' a value of five.
<P>
2.10 The Deaths Variable
<P>
The variable &quot;Deaths from Political Violence&quot; is included
as a measure of the intensity of political violence within societies.
Gurr (1970) argues that 'Deaths' is the best estimate of the magnitude
of political violence, and numerous studies of political stability
and violence have employed this measure (see, Muller and Seligson
1987; Muller, Seligson, Fu, and Midlarsky 1989).We adopt the definition
developed in the World Handbook of Political and Social Indicators
which points out that &quot;Unlike the other indicators, the number
of deaths from political violence is not an event variable, but
a body count.&quot; (Taylor and Hudson, 1975, p. 18). Many of
the events which are coded in the IPI system may involve deaths.
Certainly this would include armed attacks, military clashes,
assassinations, executions, and death squad activity. We will
also keep track of victims of riots, bombings, disappearances,
and others whose deaths are directly attributable to a domestic
political event.One of the problems with the 'deaths' variable
is illustrated by <EM>TNYTI</EM> report from the Philippines reproduced
above in section 2.4: summary scores will be given about the number
of dead between two given dates. A favorite approach of states
(and, hence, journalists) is to note the number of casualties
since some defining event which marks a renewed round of violence
(e.g., 211 people have been killed since violence erupted six
weeks ago). 
<P>
To address this problem we have included a 'Since When' field
in our coding sheet which coders use to indicate the length of
time over which the death count is aggregated. This information
will be used at a later date to estimate values for smaller time
periods. The 'Since When' field will be left blank when the number
of deaths reported are not a cumulative score since a given date.
Cumulative death scores, however, should not be reported as part
of an event record. Instead, a report of a cumulative death count
should employ only the following four fields on the coding sheet:
'Deaths', 'Since When', 'Reference', and 'Description of Event',
with the possible inclusion of 'Needs More Information' and 'Comment'.
<P>
Another difficulty in keeping track of the number of deaths from
political violence is the fact that we often receive conflicting
reports of the number of deaths. New information becomes available
as time passes. The number of deaths reported two or three days
after a riot or bombing may be significantly higher than the count
which appeared in the first report of the event immediately following
its occurrence. In this case, we update the death count for the
event to the final figure after all casualties are reported. In
some instances the updating of the number of deaths may alter
the appropriate code for the event itself. A riot with ten deaths
or fewer, for instance, is coded as a 504 on the IPI conflict
scale. A riot with eleven to ninety-nine deaths, however, is a
602. Were the death count for a riot to rise from an initial report
of eight to a final report of fifteen, the coder should not only
correct the entry in the 'deaths' field, but change the coding
of the original event as well.
<P>
Finally, it is also frequently the case that parties to a conflict
disagree as to the number of deaths attributable to a conflictual
event. We hope that in these instances our sources will provide
independent counts, or at least suggest which claim is more reliable.
If we have no way of determining from the event report which casualty
figure is more reliable, we record the more conservative number
in the 'Deaths' field, report the discrepancy in the 'Description
of Event' field, and check the 'Needs More Information' field
in the hope that we will be able to clarify the report from other
sources. 
<P>
2.11 The Coding Sheet
<P>
A copy of the IPI Coding Sheet can be can be obtained from the
authors. It is simply a hard-copy of a blank spreadsheet created
with Quattro Pro. Many of the fields are self-explanatory. The
exceptions are discussed in more detail below.
<P>
First, there is a 'Multi' field. As discussed in section 2.4,
the IPI coding scheme considers a number of episodes to contain
multiple events. Long term strikes, protests which spread to several
cities or involve many groups, and mass arrests all result in
multiple IPI event records, as we code a separate event for each
day, each group, and each location involved in a particular action.
Because some end users of the data may choose to manipulate these
episodes to lessen their impact, we identify them with an indication
in the 'Multi' field. The following are the appropriate entries
in the 'Multi' field: 0 Not a Multiple Event 1 Multiple Days (e.g.,
Teachers strike for 100 days) 2 Multiple Groups (e.g., Students,
workers, and peasants marched) 3 Multiple Locations (e.g., PRD
led protests took place in five cities) 4 Multiple Numbers (e.g.,
Over 500 were arrested) 5 More than one of the above codes applies
(e.g., Students, workers, and peasants marched for five days)
<P>
Second, there is a 'Since When' field. As discussed in section
2.10, 'Since When' is used primarily in conjunction with the 'Deaths'
variable. Journalists frequently provide summary reports of the
number of people killed since a given date. To keep a running
tally, it is important to keep track of the time period covered
by the summary report. A date in YRMODA format (i.e., year, month,
day) should appear in the 'Since When' field. In general, the
'Since When' field is not employed as part of an actual event
record. Instead, it is used to keep track of useful and necessary
information reported in our sources but not connected to identifiable
events.
<P>
Third, there is a 'Reference' field. This field enables the coder
to document the Month, Day, Year, and Page Number of the report.
When working with Reuters news reports online, coders should copy
the proper Reuters article reference (e.g., rl040586.138) into
the 'Reference' field on the IPI coding sheet.
<P>
Fourth, there is a 'Description of Event' field which is employed
to summarize the event briefly. Because we employ multiple sources,
we must be able to distinguish among multiple similar events.
The 'Description of Event' category is used for this purpose.
This field also facilitates editing of the IPI data. Should recoding
of a particular class of events be desirable or necessary in the
future, the 'Description of Event' field makes this possible.
Coders should provide the most specific information given in the
news source regarding the actor, target, date, and location of
the event in this field. Careful use of the 'Description of Event'
field will allow end users to create additional distinctions in
the data to further examine the political dynamics of a particular
case. 
<P>
Fifth, we employ a 'Needs More Information' field. This field
is particularly useful for vague reports like &quot;sustained
guerrilla activity was reported yesterday&quot; or &quot;fifty
bombings have occurred in the last month&quot;. It should be checked
any time the coder thinks that there is missing information because
the report is incomplete. There is no need to check 'Needs More
Information', however, if the nature of the event implies that
the information does not exist. For instance, if the 'actor' is
missing in a bombing report, then there is no reason to check
'Needs More Information' because no source will authoritatively
provide us with the missing information. The utility of having
this field is that it draws our attention to search other sources
for information on a given event and should help us to minimize
doublecoding. Thus, 'Needs More Information' should be checked
any time that the coder thinks a record is incomplete due to problems
with the source rather than the nature of the event. 'Needs More
Information' should be left blank when it is not used, and have
a Y in it when it is needed. 
<P>
The 'Needs More Information' field is also useful when events
are reported as likely to happen, but the source is mute as to
whether they actually happened. For example, a source may report
that military officers have been sentenced to death and never
report that the execution actually occurred. In this case, we
code the conviction and check the 'Needs More Information' field,
making a note in the 'Comment' field to look for a report of the
execution in another source. The 'Comment' field should be used
in conjunction with the 'Needs More Information' field. Whenever
'Needs More Information' is checked, an entry in the 'Comment'
field should describe what information is needed. In addition,
coders may use the 'Comment' field to direct concerns, questions,
or further explanations to the editor. Any uncertainty about the
coding of an event report should be indicated in the 'Comment'
field for future review.On the coding sheet, coders should leave
blank all fields which are not relevant for a particular event.
The one exception to this rule is the case of the deaths variable.
When no deaths occur, a zero should appear in the field.
<P>
2.12 Managing Computer Files
<P>
Data management is a crucial aspect of the IPI project. Each coder
should maintain at least two copies of all of his or her work,
one on a computer hard drive or server, and at least one on floppy
disk. Each country/year should occupy a separate quattro pro file.
The protocol for naming IPI computer files is as follows: (1)
three alphabetical digits for the country name; (2) two numeric
digits for the year; (3) three alphabetical digits for domestic.

<P>
The following are the three digit country codes for our initial
cases: 
<P>
ARGENTINA ARG 
<P>
BELGIUM BEL 
<P>
BRAZIL BRA 
<P>
CHILE CHI 
<P>
COLOMBIA COL 
<P>
GREECE GRE 
<P>
HUNGARY HUN 
<P>
INDIA NDI 
<P>
INDONESIA NDO
<P>
IRAQ RAQ 
<P>
KENYA KEN 
<P>
KOREA (SOUTH) ROK 
<P>
MEXICO MEX 
<P>
NIGERIA NIG 
<P>
PAKISTAN PAK 
<P>
PHILIPPINES PHI 
<P>
SOUTH AFRICA SAF 
<P>
SPAIN SPA 
<P>
SYRIA SYR 
<P>
TAIWAN TAI 
<P>
VENEZUELA VEN 
<P>
ZAIRE ZAI
<P>
ZIMBABWE ZIM
<P>
Thus, the IPI coding for Nigeria 1985 would be located in a file
called NIG85DOM.WB1. The coding for Argentina 1982 would be named
ARG82DOM.WB1. Proper naming of files is essential to allow us
to distinguish between cases, and also between coded and edited
data. Edited data is saved with a different nomenclature. 
<P>
3. ACTOR AND TARGET CODES
<P>
3.1 Selecting Actor and Target CodesOne of the primary goals of
the IPI project is to create data which allows analysts to examine
the behavior of particular societal groups and their interactions
with other social groups and the state. As a result, the IPI coding
scheme provides a large number of actor and target codes so that
groups can be identified and tracked with a greater degree of
precision than has been previously possible. Because of this increased
precision and the plethora of available actor and target codes,
it is often incumbent upon coders to make difficult decisions.
Many actors and targets belong to more than one category. In this
case, the coder must determine which group the actor or target
is representing in the event in question. For instance, if a person
is arrested who is both a labor union leader and a member of the
communist party, the coder must determine if this individual is
being arrested because he or she is a union leader or because
he or she is a communist party member. We hope that this information
can be gleaned either from the report itself or from the context
of other reports. Similarly, when a former President makes political
statements or undertakes political actions, the coder must determine
if the ex-President is acting as a former leader or as a member
of his or her political party. 
<P>
It is also the responsibility of coders to translate journalistically
assigned group identities to the appropriate IPI actor or target
codes. The word &quot;farmers&quot;, for instance, could conceivably
be most accurately captured by three different IPI codes. Farmers
may be peasants (50); they may be landed elites (65); or they
may be a business sector (52). The context of the report should
suggest which code is most appropriate in a given instance. In
all cases, coders must report the precise description of the actor
and the target as given in the news source in the 'Description
of Event' field, including proper names when appropriate. This
allows end users of the data to alter the coding of specific groups
with relative ease, or to create new actor and target groups.

<P>
The IPI Actor and Target Codes are listed in the appendix. The
scheme is adopted from the system developed for identifying sub-national
actors in Ted Robert Gurr's &quot;Minorities at Risk Project&quot;
(1993) and John Davies' GEDS project (1993). We use the standard
ICPSR country code to identify the state and append a two or three
digit code to identify specific actors. Coders then assign specific
codes for major ethnic populations, dissident organizations, and
major political parties that appear in that case (e.g., the FARC
in Colombia; the Radicals in Argentina) working within the framework
established below (see section 3.11). In the following sections
we provide a more detailed discussion of some of the generic and
case specific actor and target codes. 3.2 The State (10-24)
<P>
In the IPI system, the state is disaggregated into the following
governmental actors: (10.1) National Executive, (10.2) Head of
State (if distinct from the national executive) (11) High level
officials (Cabinet members), (12) Elected representatives (legislature),
(13) the Judicial Branch, (14) Regional/state governments, (15)
City/village governments, (16) Military and armed forces, (17)
Paramilitary forces (i.e., death squads), (18) Police forces,
(19) Deposed leaders, (20) Former leaders, (21) State enterprises,
(22) Dissident military faction, (23) Former military leaders,
(24) Other. Government workers are treated separately as social
actors. This coding scheme allows us the flexibility to record
conflict and cooperation within the state apparatus. We can code,
for instance, criticism of the executive in a democratic state
by the ruling party, or a purge of the army by a ruling military
junta. If there is a coup attempt by junior officers, the event
can be coded with actor as military/armed forces (16) and the
target as national executive (10.1). Further if the executive
purges his or her cabinet or junta, the actor is 10.1 and the
target is 11. When students clash with the police, the actors
and targets are 49 and 18.Actor and Target Code 10 represents
the State as a whole. When dissident groups, social actors, or
ethnic populations conflict or cooperate with the government,
10 is usually the appropriate actor/target code to represent the
state. Only when the actions of a societal group are focused specifically
on particular governmental actors (e.g., a city government (15),
the military (16), or the legislature (12)), should coders employ
the more specific codes. In addition, when actions of the state
toward the population or particular groups within it are statements
or acts of overall government policy, 10 is the appropriate actor
code. For instance, if troops are placed in an area of heavy guerrilla
activity, this is an act representing mobilization of resources
by the state against the guerrillas (actor code 10, target code
45, conflict code 304). If the President signs a truce with a
guerrilla group, he or she is representing the State, and the
cooperation occurs between 10 and the guerrillas. If, however,
the Minister of the Interior expresses an opinion that increased
action should be taken against guerrillas, this would be a threat
from 11 (high level officials) to 45, as it is unclear that this
is the policy of the government as a whole.
<P>
In states with coalition governments, we additionally seek to
distinguish between intra-coalitional conflict and cooperation
and extra-coalitional conflict and cooperation. In these states
(for instance, Belgium), we use actor and target code 11 (high
level officials, cabinet members) to represent the ruling coalition.
Thus, if leaders of the coalition parties threaten the national
executive, the actor code is 11 and the target code is 10.1. 
<P>
3.3 Mass/Elite Distinction
<P>
Just as we track conflict and cooperation within the government,
we also make it possible to track interactions within certain
societal groups by distinguishing between mass and elite actors
and targets. We define elites as those individuals who control
significant social resources. Whenever actions are either directed
at or undertaken by only a portion of a group (i.e., only the
leaders or only the masses), we append a decimal to the group
code to reflect the more specific nature of the actor or target
group as indicated on the Actor/Target Scale. This distinction
between mass and elite actors allows users to investigate differences
in the dynamics of intra-elite and mass interactions and to capture
conflict between masses and elites of the same group. 
<P>
3.4 Unspecified Actors and Targets (9, 45, 75, 89, 99)
<P>
There are a number of different codes for unspecified actors and
targets. The most general code is 09, which represents the general
population. Only in very rare cases should 09 serve as an actor.
Most of the time, the code for general population is only used
as a target; it is a group which is acted upon, usually by the
state as it imposes or relaxes general repressive measures. The
more common codes for unspecified actors and targets are unspecified
guerrillas (45), unspecified social actors (75), and unspecified
political parties (89). The code for unspecified guerrillas should
be employed either when the report defines the actor or target
as guerrillas but is not more specific, or when the action undertaken
suggests that the actor was clearly a dissident group which has
adopted armed struggle as its mode of operation. For instance,
if unknown assailants ambush soldiers and a group of guerrillas
is active in the case being coded, we can make the assumption
that a violent dissident group was the actor and we can code the
actor as 45.
<P>
The code for unspecified social actors (75) is commonly used in
cases of reports of demonstrations, mass arrests, and other such
actions which affect large numbers of people who may not share
identifiable characteristics other than their common action. It
is also used to represent conglomerations of groups. For instance,
as discussed in section 2.3, when several groups coordinate their
activities we code cooperation from each group to all of the other
groups using the unspecified social actors code (75).The code
for unspecified political parties (89) should be employed when
neither the party nor its political leanings can be identified.
For instance, if a known political party is reported to have formed
an alliance with five other parties which are not named in the
report, we code cooperation between the known party and unspecified
political parties (89). Coders should note, however, that we do
provide codes for unspecified leftists (87), and unspecified rightists
(88), so when the political leanings of a party can be determined,
these are more appropriate codes than is 89. In addition, we employ
the various unspecified codes when a party, social group, or dissident
group is identified in a report, but does not qualify for an IPI
case specific code (see section 3.11). Dissident groups which
exist for less than a year, for instance, should be coded as unspecified
guerrillas (45). Similarly, social actors which do not appear
on the generic list should be coded as 75.
<P>
Actor and Target code 99 indicates missing information. This code
should not be used to represent unspecified actors and targets,
but rather unknown actors and targets. It will be common, for
instance, for the actor in a bombing to be coded as 99.
<P>
3.5 Business and Labor (51, 52, 64, 54, 55, 56)
<P>
There are three codes in the IPI coding scheme to represent business,
and three to represent labor. The business codes are distinguished
from one another as follows. Actor/Target code 51 represents organized
business groups. The code for business sector (52) is used to
represent a group of business people employed in the same sector
working together. If a report states, for instance, that &quot;bankers
declared support for the opposition economic program and criticized
government handling of the economy&quot;, the bankers would be
coded as a business sector (52). 64 is the default business code.
If a report simply states, &quot;labor, business, and the government
met today to discuss inflation&quot;, we code business as 64.
The three labor/worker codes are distinguished as follows. Labor
confederations are coded as 54. Labor confederations are large
organizations which bring together workers from many regions and
many occupations into one umbrella organization. An example is
the CGT in Argentina. Code 55 is used to represent individual
labor unions. Code 56 is employed to represent unspecified or
unaffiliated workers. If it is unclear than an event involving
workers was organized by a labor union (e.g., &quot;workers and
students marched on the capital to protest price increases&quot;),
we use the code 56. Code 53, however, represents government workers.
Thus, when the workers are exclusively state employees, 53 is
a more appropriate code for unorganized workers than is 56.
<P>
3.6 Multinational Groups and Foreign Nationals
<P>
When multinational groups and foreigners become involved in the
political activities of a state from bases within the state, they
are codable as actors and targets under the IPI scheme. It is
important to recognize, however, that only the local actions of
these groups are domestic political events. Foreign media (47),
for instance, can act and be acted upon in a domestic political
context. The Roman Catholic Church (57), human rights groups (62),
and public interest groups following such issues as environmental
policy (63), are active in many states and can be coded as domestic
actors in all of those states. The international actions of these
groups, however, are not codable as domestic events. A pronouncement
by the Pope can not be a domestic political event, but a pronouncement
by a local Bishop can. Reports by Amnesty International issued
in Washington, DC, are not domestic political events, but statements
by the local Amnesty office are.The code for foreign nationals
(70) should be employed only when an individual is acting or being
acted upon because he or she is a foreign national. For instance,
if Chileans living in Argentina protest Argentine policy toward
the Beagle Islands, we code a domestic political event with foreign
nationals (70) as the actor. But if a member of a labor union
who happens to be a foreign national launches a protest, we code
the actor as labor union (55), not foreign nationals.3.7 Leftists
and Rightists (73, 74, 87, 88)
<P>
The IPI system provides Actor and Target codes to represent leftists
and rightists both in the social actors category and in the political
parties category. As mentioned above in section 3.4, codes 87
and 88 represent unspecified political parties with left and right
leanings respectively. Often, however, actors and targets will
be identified in news reports as &quot;leftists&quot; or &quot;rightists&quot;
when they are not representing organized political parties. In
these cases, we employ the codes 73 and 74. Consider the following
statement: The President blamed leftists for inciting the student
riots.In this case, we code the actor as 10.1 and the target of
the negative statement (101, conflict scale) as 73. Similarly,
a report like &quot;Right wing sympathizers protested in front
of the Ministry of Defense to express their anger at proposed
defense cuts,&quot; would be coded with 74 as the actor and 10
as the target. 3.8 Religious Groups (57, 58, 59, 60, 71)
<P>
There are a number of IPI actor and target codes which represent
various religious groups. A religious group code can only be employed
as the actor in an event when the action is taken by official
leaders or members of the faith acting in a manner commensurate
with a sanctioned position of the religion. Thus, if members of
the Catholic Church act violently toward abortion providers, we
can not code 57 as the actor because the Catholic Church does
not sanction violent action. These actors are not a Catholic social
group, but a political dissident group. Similarly, if Islamic
extremists blow up a building or take hostages, we code their
actions as the actions of a political dissident group.
<P>
Any time members of a faith are targeted because of their religious
faith, religious group codes can be employed as targets. If Christians
are targeted for repression because they are Christian, we can
code 58 as the target. In many cases, however, religion and ethnicity
will be correlated with one another. When a group which shares
a religion is truly targeted for its ethnicity, the ethnic group
code is more appropriate. Whenever there is a question as to whether
a group should be defined by its ethnicity or its religion, the
default should be ethnicity.In addition, conflict between factions
within a given religious faith (e.g., Sunni v. Shiite) is captured
through the use of ethnic group codes. Thus, when a particular
sect takes action, the appropriate code is often an ethnic group
code.
<P>
3.9 Socially Influential Individuals (66)
<P>
Actor and target code 66 captures the actions of socially influential
individuals, people who command attention due to the respect,
fame, and popularity they have achieved in their chosen fields.
Certainly, if leading economists comment on an economic plan,
the actor code might be 66. Also codable in this field, however,
would be famous musicians, novelists, actors, and sports players
whose political activities receive special attention. For instance,
in Argentina in 1986, immediately before the World Cup, the manager
of the Argentine national soccer team makes political statements.
Because we believe his statements are socially influential because
of his high stature in his field, the actor is coded as 66. Similarly,
we often code the actions of family members of high level government
officials (e.g., the wife or brother of the President) with actor
code 66.3.10 Supranational Organizations (95)
<P>
IPI Actor and Target Code 95 represents Supranational Organizations.
The only cases for which this code is currently relevant are those
which are members of the European Community. Interactions between
the European states and the European Community are international
events and are not codable under the IPI system. Actions of domestic
political groups toward the EC, however, are codable as domestic
political events. Thus, if Belgian citizens protest against EC
nuclear policy in Brussels, the actor is 75 and the target is
95.3.11 Case Specific CodesIn addition to the generic actor and
target codes which are applicable to all states, the IPI coding
scheme tracks the most important political groups specific to
each of our cases. First, we assign specific codes to major ethnic
populations within each of our cases (codes 01-08). We begin with
the groups identified by Ted Robert Gurr's &quot;Minorities at
Risk Project&quot; (1993) and add additional groups as necessary.
Second, we assign specific codes to dissident groups (codes 25-44)
which remain politically active in a state for a minimum of one
year. Third, we designate codes for the major political parties
(76-85) in each state. We define major parties as those which
consistently receive more than five percent of the vote nationally,
or which frequently receive a winning plurality of the vote regionally.
Currently assigned case specific codes can be located in the appendix.
In addition, brief descriptions of coding decisions specific to
each case are available from the Principal Investigators upon
request.
<P>
3.12 Distinguishing Among Ethnic Groups, Social Groups, and Dissident
Groups
<P>
It is important for coders to distinguish between political dissident
groups and social or ethnic groups. Some of these distinctions
(between religious groups and ethnic groups and between religious
groups and dissident groups) were discussed above in section 3.8.
Similar issues arise in distinguishing the instances in which
it is appropriate to employ an ethnic group code and when it is
appropriate to employ a dissident group code. Ethnic groups are
not organized entities; they are merely collectivities which share
common characteristics. While members of such groups may be targeted
due to these characteristics, it is rare that action erupts spontaneously
from these groups. As a result, the IPI ethnic group codes are
more often used to identify targets than actors. When groups or
organizations emerge to pursue the interests of ethnic groups
politically, these groups are functioning as political dissident
groups. Separatist movements, for instance, are political dissident
groups, not ethnic groups. Thus, in the IPI system, we code the
group or organization which instigates action as the actor, not
the broader social grouping into which members of the group may
fall. 
<P>
3.13 Coding Factions and Coalitions
<P>
As factions and splinter groups begin to act independently from
their parent organization, we create new codes for them by appending
a decimal to the code of their parent organization. For example,
the FARC is actor/target code 26 in Colombia. FARC factions are
coded as 26.1, 26.2, etc. If a faction or splinter group survives
as an independent group for a year or more, it receives its own
actor/target code. For example, in Colombia, in 1984, a faction
of the FARC rejected a cease fire agreement and chose to continue
the armed struggle. This group continued to act independently
for more than a year under a new name: the Ricardo Franco Front.
The Ricardo Franco Front is thus assigned its own actor/target
code in Colombia (28). Similarly, when coalitions form which act
consistently for their member groups for a year or more, they
too receive independent actor/target codes. For instance, the
Argentine multipartidaria, formed in 1981 as an alliance of five
political parties (the Peronists, the Radicals, the Christian
Democrats, the Desarrollistas, sand the Partido Intransigente)
is actor/target 80. When questions arise, coders should err in
the direction of proliferating the number of actors. Users of
the data can always aggregate actors as they see fit, but they
cannot disaggregate beyond what the data provide. It is important
to note as well that when members of any dissident group, political
party, or social group renounce their ties to that group, they
can no longer be coded as speaking for the group in question.
For example, if a few members of a dissident group surrender,
renounce their ties to the group, and inform the state about the
group's activities, we code these former members as actors in
cooperative events with the government and conflictual events
with the dissident group. They are no longer coded as members
of the dissident group, however. Instead, we code their actions
as the actions of unspecified guerrillas (45). Similarly, if a
few members of a political party resign from the party and criticize
its activities, we do not code their actions as emanating from
that party itself. Instead, the appropriate actor code is unspecified
parties (89) or unspecified leftists or rightists (87 or 88).

<P>
4. THE CONFLICT SCALE
<P>
4.1 Introduction to the Conflict Scale
<P>
The IPI conflict scale is designed to facilitate the measurement
of the volume and intensity of political conflict within a domestic
polity. Events are assigned values on a one to ten scale according
to their severity. In accordance with accepted scholarly practice,
weights capturing the relative intensity of each category of political
events will be developed by polling a panel of experts (see, for
example, Goldstein 1991, and Moore and Lindstrom 1994). These
weights can then be used to develop indices of the intensity of
conflict and cooperation.In general, 100 and 200 level events
represent verbal conflict and 300-1000 level events are conflictual
actions. 100 level verbal conflict implies no threat of action,
whereas 200 level verbal conflict involves an explicit or an implicit
threat to the target. 300 level events are always nonviolent,
whereas actions starting at the 400 level may involve violence.
In the following sections we discuss further rules for coding
particular conflictual events. Coders should additionally refer
to the scale itself in the appendix for helpful examples.
<P>
4.2 Press Criticism of Government (101)
<P>
We code press criticism of the government only when it is reported
by a secondary source, not when it appears in the source we are
coding. In other words, if we are coding a local publication,
we do not code a verbal conflict event from the media to the state
for every critical editorial which appears. If another source
writes a news report about a critical local editorial, however,
it is appropriate to code a verbal conflict event (101) with the
media (46) as actor and the state as target. A similar rule applies
to letters to the editor. They may be coded as verbal conflict,
but only when they are discussed in a secondary source.
<P>
4.3 Rumors of Threatening Action (201)
<P>
When our sources report rumors of threatening action, we code
them as actual threats (201). We are interested in measuring the
perceived threat by the target, and we believe that reports of
rumors of threatening action are perceived by the target as threatening.
When a rumor of a potential coup d'etat surfaces, for instance,
we consider it to be as threatening to the executive as an explicit
threat would be. 
<P>
4.4 Strikes, Protests &amp; Riots (301, 302, 401, 404, 504, 602,
or 701)
<P>
These three types of events raise a number of coding ambiguities
that must be addressed. One difficulty concerns long-term strikes
or protests (i.e., those that persist for substantial periods
of time such as the anti-Marcos demonstrations in the Philippines
prior to the `People Power Revolution' or the pro-Democracy demonstrations
that preceded the Tianenmen Square crackdown in China). The problem
is that once these events become commonplace, they are no longer
treated as newsworthy by the press unless something unusual (like
violence) occurs. This can lead to reports like the following
fictitious report: Violence broke out in the 2 week old stand-off
between low-paid bureaucrats in the capital and the armed forces.
Three people were reported injured in the clash.Sometimes, this
will be the first and only report of the event. In these cases,
we do not want to code fourteen days of violent strike activity
(401s and 402s on the conflict scale), but rather fourteen 302s
and one 402 (note that since the report does not indicate that
the bureaucrats were violent, we assume that only the police were).
It is also important to note that in cases where a peaceful demonstration
or strike erupts in violence, we do not code both a demonstration
and a riot for the same group on the same day. We count the entire
day at the highest level of conflict reached. 
<P>
A second issue with these types of events concerns the fact that
we distinguish among major and minor types. These are generally
distinguished by the damage--typically to human beings--done (e.g.,
riots) or the number of cities that are involved (e.g., strikes).
These distinctions are noted in the scales below. The distinction
between local and national strikes (i.e. 302 v. 404, conflict
scale), for instance, is whether a majority of major urban areas
are involved. Thus, a strike that occurs in three of the ten cities
in country X, but involves only salaried government employees,
is three 302s (conflict scale). If the employees are joined by
the armed forces in the capital, but not elsewhere, we code four
302s. A national strike involving salaried government employees
and the armed forces would instead be coded as two 404's. One
could make the case that protests and demonstrations should also
be coded with respect to the verbal conflict that they entail.
As discussed above in section 2.4, however, we do not code these
events as verbal conflict (i.e., the verbal conflict is implied
by coding it as a 302 or 402). 
<P>
Lastly, the coder must be able to differentiate between those
strike demands that are economic in nature and those that are
political. Strikes for higher wages are economic and not political
events. As such, they are not codable under our scheme. We consider
strikes against the minimum wage (which is a government policy)
or the length of the work week, however, to involve political
demands and to be codable as political events. We also consider
all strikes by state employees to be codable IPI events.When a
long term strike or demonstration ends, we stop coding conflictual
events. In some cases, however, it is appropriate to code cooperative
events as well which correspond with the cessation of hostilities.
If active negotiations and/or a specific agreement precedes or
corresponds with the end of the strike or demonstration, we code
cooperative events between the parties. If, however, the strike
or demonstration ends with no active resolution, we do not code
cooperative events. 
<P>
4.5 Executive Adjustment, Political Resignations (306, 301)
<P>
The key difficulty in coding these events is deciding how many
events to code. A single executive adjustment can involve the
replacement of several cabinet ministers. We adopt the rule that
only one executive adjustment can occur per day. Thus, if the
entire cabinet is asked to resign at once, we code one executive
adjustment (306), just as we do if only the finance minister is
asked to resign. If one minister is replaced on Monday and another
on Thursday, however, two executive adjustments are coded. Similarly,
if ten legislators resign together in protest, we code one political
resignation (301). It is also important to note that a single
executive adjustment is often a two step process. Many times,
a cabinet minister is asked to resign on one date, and his or
her replacement is named at a later date. The firing of the original
minister and the hiring of a new one constitute a single executive
adjustment; only one event (306) should be coded. We recognize
the date of the firing of the original minister as the date of
the executive adjustment in our coding. Finally, we are interested
only in political conflict events. As a result, changes in personnel
attributable to illness, natural death, retirement, pursuit of
other political office, or other non-political explanations are
not coded under the IPI rules.
<P>
Occasionally it is unclear when a member or members of a Cabinet
resign whether it is appropriate to code political resignations
or an executive adjustment. In other words, there is sometimes
ambiguity as to whether the politicians in question are resigning
of their own accord due to policy differences or have been asked
by the national executive to leave. When it is impossible from
our sources to determine which of these is the case, we code an
executive adjustment with the executive (10.1) as actor and cabinet
members (11) as the target. 
<P>
4.6 Censorship (303, 406, 501)
<P>
There are three different codes for censorship on the IPI conflict
scale at the three hundred, four hundred, and five hundred levels.
Individual acts of censorship are coded at 303. At this level,
each individual act of censorship is a separate event. Closing
two newspapers, for example, is two 303s. When a larger number
of publications of a given type (e.g., Jewish books in Nazi Germany;
all leftist newspapers) are closed, however, we consider this
&quot;widespread media censorship&quot; and record a single 406.
Complete censorship, involving a total lack of uncontrolled media
access, is a single 501. 4.7 Mobilizing Resources (304, 502)
<P>
The conflict scale offers two codes for mobilizing resources and
increasing power: 304 and 502. The difference between the two
codes is whether the actions taken are legal or illegal. When
groups use legal means to increase their power in relation to
their adversaries, we code the events at the three hundred level.
Using illegal means to increase power, however, displays a higher
degree of conflict. As a result, illegal means of mobilizing resources
are coded at the five hundred level. Coders should consult the
conflict scale for examples of activities which fall into each
of these categories. 
<P>
4.8 Violent Political Acts (405, 505, 604)
<P>
Events 405, 505, and 604 on the conflict scale all represent violent
political acts of differing intensities. Code 405 should be used
for acts of intimidation which last less than twenty-four hours.
Some examples are brief abductions or beatings. In general, these
events are scare tactics or mild shows of force. Code 505 represents
acts which involve a higher level of violence, but which do not
involve deaths. Bombings, hijackings, kidnappings, and shootings
which do not result in injury or deaths belong at this level.
Violent acts which result in casualties should be coded at the
600 level (604). Whenever these distinctions are unclear in a
report, we assume the less intense action. 
<P>
For instance, if it is unclear whether an abduction lasted more
or less than twenty-four hours, we code a 405. When there are
no indications of injuries as a result of a bombing, we code a
505. Finally, when the same person is kidnapped and killed on
the same day, this constitutes only one violent political act
(a 604). If the two actions occur on separate days, however, we
record both a kidnapping (505) and an armed attack (604) with
one death from political violence. 
<P>
4.9 Mass Arrests and Political Convictions (503, 606)
<P>
Event 503 on the conflict scale represents the arrest (among other
things) of dissidents. We do not, however, code a separate event
for the arrest of each individual. Rather, we code a separate
event for every group of 100 people who are arrested, and a separate
event for each leader arrested. This decision reflects our belief
that it constitutes a higher level of conflict to arrest the leader
of an organization than to arrest rank and file members. The following
scale applies: 1-100 arrests 1 503 101-200 arrests 2 503's 201-300
arrests 3 503's 301-400 arrests 4 503's 401-500 arrests 5 503'sand
so on. For example, if seven members of a dissident group are
arrested, it is a single 503. However, if 750 'rank and file'
and ten labor union leaders are arrested, it is coded as eight
503s (target=55.2) and ten 503s (target=55.1).
<P>
Consider the following report from Latin America Weekly Report,
April 2, 1982: The CGT-organized trade union demonstrations against
the government on 30 March were met with police repression. In
Buenos Aires running battles between demonstrators and police
went on for four hours; over 400 arrests were reported. . . .
Saul Ubaldini, the CGT general secretary, was arrested as he tried
to lead the march on the Casa Rosada in Buenos Aires (page 12).
In this case, we code five 503s (target 54) and one 503 (target
54.1). We use mass and elite subcodes only when the report specifically
makes that distinction. It is important to keep in mind as well
that if several groups are targeted, or if arrests occur in many
cities or on many days, each of these is a separate event regardless
of the total number of people arrested.
<P>
When those arrested are actually convicted of political crimes,
we code a separate event for the conviction (606 on the conflict
scale). We use a similar rule for calculating the number of events.
Each leader convicted of a political crime constitutes a separate
event. When a mass of people are convicted at once, we count one
conviction for every 100 people affected. Occasionally, our sources
will fail to report a political conviction, but will report a
sentencing. In this instance, we code a political conviction on
the basis of the sentencing report and check 'Needs More Information'
in the hope of clarifying the conviction date from other sources.In
some cases, the very act of bringing certain people to trial is
politically significant. Particularly when former leaders or high
ranking opposition members are tried, the trial itself is newsworthy
and politically significant. When the beginning of a political
trial is reported in our sources, we code the act of bringing
an opponent to trial as a 304 on the conflict scale, a legal action
to mobilize resources.
<P>
4.10 Bombings (505, 604)
<P>
There are two bomb codes (both on the conflict scale): 505 and
604. The distinction between these two codes is whether human
beings are injured or killed as a consequence of the bombing.
Attempted or failed bombings (i.e., defused bombs), sabotage bombings
(economic targets such as electric pylons), etc., should be coded
505. Bombs (land mines, car bombs, etc.) that kill or injure people
should be coded 604. Because the state claims a monopoly over
coercive behavior, most bombings should list the state (10) as
the target (see section 2.2). The exception is a well documented
bombing of a political, social, ethnic, or religious group, in
which case that group is coded as the target. A bombing of the
headquarters of a political party or labor union, of a place of
worship, or of a group of businesses owned by members of a particular
ethnic group targets the group in question, not the state.
<P>
4.11 Assassinations (605, 702)
<P>
Codes 605 and 702 on the conflict scale are used to record unsuccessful
and successful assassinations. We follow Taylor and Jodice's World
Handbook of Political Events in defining potential assassination
targets. Taylor and Jodice state: Political Assassination . .
. is a politically motivated murder or attempted murder . . .
of a national leader, a high government official, or a politician.
Among the national leaders included are chiefs of state, heads
of government, ministers, legislators, judges and high-ranking
civil servants and military officers. In addition to national
leaders, state, province or district governors, mayors of large
cities, and newspaper editors are included. Nationally prominent
politicians who are not holding office (e.g., Dr. Martin Luther
King and Robert F. Kennedy) were also included (1975).
<P>
Political murders and attempted murders of citizens who do not
fall into these target categories (e.g., newspaper reporters,
local trade union leaders, etc.) should instead be coded as 604
(if the target dies) or 505 (if the attempted murder is unsuccessful).
<P>
4.12 Reports of Disappearances (608)
<P>
Disappearances are coded as deaths from political violence in
this project. A report which states, &quot;Twelve human rights
activists disappeared on Tuesday. Death squad activity is suspected.&quot;
is coded as twelve deaths from political violence and one report
of disappearances (608 on the conflict scale). We treat summary
reports of disappearances much like we do those of deaths (see
section 2.10); we record the number of disappeared in the 'Deaths'
field and complete the 'Since When' field to indicate the period
over which the count is aggregated. For summary reports, however,
it is not necessary to code a 608, as summary reports are not
in themselves events.
<P>
4.13 Breaking a Truce, Breaking a Peace Treaty (603, 707)
<P>
These two events represent the abrogation of an agreement and
the renewal of hostilities. The distinction between a peace treaty
and a truce is discussed further in section 5.7. When a party
to a truce or a peace treaty breaks the treaty by committing acts
of violence, we code a 603 or 707 on the conflict scale for the
act of breaking the agreement, and we additionally code the acts
of violence at their appropriate level on the conflict scale.
<P>
4.14 Attacks, Clashes, and Battles (604, 704, 708, 806, 1002)
<P>
There are several different codes on the IPI conflict scale representing
armed action. The least severe is the armed attack (604; 708).
Armed attacks occur when one armed group or individual fires on
an unarmed group or individual or when two armed groups exchange
fire in an unorganized fashion (i.e., gunmen fire at police station
and police fire back). An armed attack can involve no more than
fifteen deaths at the 604 level, and no more than thirty deaths
at the 708 level. If more than thirty deaths result from an armed
action, the event should be coded either as a military clash (704)
or as a massacre of civilians (806). A military clash (704) is
an organized armed action involving groups which have adopted
armed struggle as their accepted course of action. Common examples
include battles between guerrilla groups and the army or between
rival ethnic groups. 
<P>
It is important to recognize that the distinction between an armed
attack and a military clash is the organized nature of a military
clash, not the number of deaths. A military clash need not involve
an extensive number of deaths to be coded as such, but military
clashes are always reciprocal events involving at least two actors.A
massacre of civilians (806) is an attack by an armed group (either
the state or another actor) on an unarmed group which results
in more than thirty deaths. If fewer than thirty are killed as
a result of the attack, the event should be coded as a 708 or
a 604.Finally, event 1002 represents a major battle in which both
sides defend territory with conventional military tactics and
weaponry. We expect that events falling in this category will
involve extensive casualties.
<P>
4.15 Occupation of Territory (808, 901)
<P>
In order for an event to qualify as &quot;rebel occupation of
territory&quot; (808) , both sides must recognize that the rebels
control the territory. The rebels must display widely recognized
symbols of control and take up defensive positions. Simply hiding
successfully from the government in an area without being dislodged
does not qualify as occupation of territory. The requirements
for 901, &quot;rebels setting up rival government&quot;, are even
more strict. In this case, the rebels must be able to administer
the territory (i.e., to provide services and to monopolize the
use of coercion). For both events 808 and 901 we code only the
first day. If the rebels lose control of the occupied territory,
we code the last day and complete the 'Since When' field. 4.16
Coups d'Etat (608, 705, 706, 803, 904, 1003)
<P>
We have several codes for unsuccessful attempted coups d'etat
and successful coups d'etat. The appropriate code depends on whether
the coup is successful, the degree of violence involved, and the
implications of the coup. More violent coups are more conflictual
than less violent coups, and coups which actually alter the structure
of government constitute a higher level of political conflict
than those which simply serve to rotate rulers. Event 904 (successful
violent coup which alters the structure of government) and event
1003 (successful violent coup and purge) are distinguished by
the systematic, violent elimination of the former leadership which
occurs in the latter case.
<P>
5 THE COOPERATION SCALE
<P>
5.1 Introduction to the Cooperation Scale
<P>
The IPI cooperation scale is designed to be used in conjunction
with the conflict scale. The two scales are parallel in many respects.
Together they allow us to measure the underlying level of conflict
within a society. The key to measuring cooperation on our scale
is understanding the historic relationship among groups, individuals,
and the state. Cooperative actions among parties to longstanding,
violent disputes are unusual and important events. Routine cooperation
among actors with established patterns of cooperative behavior,
on the other hand, may not reflect changes in the underlying level
of political conflict. Thus similar events between parties with
different historic relationships have a different impact on their
own societies and are coded at different levels on our cooperation
scale. We do not code routine cooperative behavior.
<P>
In the normal process of government, there are groups which have
a tendency to work together, often because they pursue similar
interests. For instance, cooperation between students and educators,
or leftist political parties and labor unions, generally has no
impact on political behavior in the society. We assume that routine
cooperative behavior represents a harmony of interests rather
than policy coordination, and as such is not what we intend to
capture with the IPI cooperation scale. We do, however, code cooperative
actions between groups which have not consistently cooperated
in the past, as this type of policy coordination is a measure
of the underlying level of conflict in a domestic polity. The
largest impact on societal relations occurs when groups which
are involved in a protracted social conflict cooperate with one
another. 
<P>
Protracted social conflict (see Azar 1990 for a detailed discussion
of the nature of the concept) refers to ongoing and seemingly
unresolvable conflict within societies over both material and
identity issues. At the broadest level, Azar (1990) argues that
a combination of concerns over individual and community identity
and security, distributive social justice and material interests
drive most incidents of protracted social conflict. The existence
of significant cleavages along identity or ideological lines and
the willingness of groups to resort to violence in pursuit of
their interests complicates conflict management. For parties involved
in longstanding, often violent, disputes with interests in strong
opposition to one another, minor actions may represent major breakthroughs.
This intensity is reflected in our scales. 
<P>
Similar events appear at different levels on the IPI cooperation
scale depending on the historic relationship between the actor
and target. For example, a meeting for negotiations between groups
which are not involved in a protracted conflict is a 203 on the
IPI cooperation scale, whereas a meeting for negotiations between
parties to an ongoing conflict is a 505.
<P>
5.2 Promises (201, 202)
<P>
There are two important issues to keep in mind in determining
if an event report is codable as a promise (201, 202). First,
promises are verbal events only. Substantive actions carrying
out promises are coded at a higher level on the scale (e.g., 303,
603). Second, to be codable, promises must be concrete. Vague
references to possible future action do not constitute promises
in this coding system. Consider, for example, the following report
from Latin America Weekly Report, January 8, 1982, on a major
speech by General Leopoldo Galtieri, President of Argentina, to
the nation: A traditional theme of military speeches over the
past six years was also present-- the nod in the direction of
an eventual return to civilian rule (p. 1). Such a vague reference
is not codable as a promise. But the following report, from Latin
America Weekly Report, June 4, 1982, is codable as a promise:
Argentine interior minister, General Alfredo Saint Jean, says
draft law legalizing activities of political parties will be released
as promised, by the end of June (p. 12).
<P>
5.3 Relaxation of Repression (301, 401, 501, 601)
<P>
Relaxation of repression appears on the cooperation scale at four
different levels: 301, 401, 501, and 601, depending on its scope
(who is affected), and the severity of the sanctions which are
ending. Coders should refer to the cooperation scale for examples
of events which should be coded at each level. Frequently, in
reporting the removal of government sanctions, journalists will
also note how long they have been in effect. In this case, coders
should be careful to code not only the removal or reduction in
sanctions but also their institution.
<P>
Sometimes it is unclear whom to code as the target of a reduction
in government sanctions. A reduction in censorship, for instance,
impacts the media most directly, but clearly has an effect on
the general population as a whole. For the purposes of this project,
we code as target the group most directly affected by a reduction
in sanctions. When it is unclear who is most directly affected,
we code either unspecified social actors (75) or general population
(09) as the target of the action. Thus, a reduction in censorship
is coded with the state (10) as the actor and the media (46 or
47) as the target. Reopening universities is coded as an action
by the state (10) directed at students (49). Legalizing political
parties is coded as a cooperative action toward unspecified parties
(89). We code the release of hundreds of unspecified political
prisoners with the state (10) as actor and unspecified social
actors (75) as target. The target of an event such as &quot;ending
a regional state of emergency&quot; is (09), general population.
As usual, each actor/target combination represents a distinct
event.
<P>
5.4 Political Reform or Accommodation (302, 603)
<P>
There are two codes on the IPI cooperation scale for political
reform, 302 and 603. The difference between the two is the scope
and intensity of the reform undertaken. A 300 level reform action
is a minor concession designed to lessen conflict, for instance,
a minor increase in wages following a devaluation. Reforms which
change the political environment, however, are coded at the 600
level. These include land reforms, the institution of a comprehensive
system of new social welfare programs, and other such similar
actions. Coders should consult the cooperation scale for further
examples. These two codes (302 and 603) represent the implementation
of these reforms. Promising reforms is coded at a lower level,
either 201 or 202. 
<P>
5.5 Talks, Agreements to Talk (203, 402, 505)
<P>
As mentioned above, there are two codes for holding talks or negotiations:
203 for talks between parties not involved in a protracted social
conflict, and 505 for groups which are engaged in protracted conflict.
In either case, we code one event per day. We have an additional
code, however, which is relevant for parties engaged in a protracted
social conflict. 402 on the cooperation scale represents either
an agreement to talk between these groups, or reports of secret
talks. Because we believe that public negotiations represent a
higher degree of accommodation/cooperation than do secret negotiations,
we code them at a higher level, but reports of secret negotiations,
especially when they are persistent, are also significant. Each
report of secret negotiations or dialogue through third parties
is coded as a 402 with each party to the negotiations as actor
and as target (i.e., there are two events in a two party negotiation).
If after a report of such negotiations, one or both groups deny
that the negotiations occurred, we code the denial(s) as verbal
conflict (101). The original report, however, remains a 402.
<P>
5.6 Surrenders, Releases of Prisoners or Hostages (501, 503, 507,
601)
<P>
The main issue to be aware of in coding these events is the number
of cooperative events to record. Just as we do not code a separate
event for the arrest of each individual (see section 4.9), we
do not record a separate event for the release of each individual.
Again, we code one event for each 100 rank and file member of
a group and one event for each group leader who is released or
who surrenders. Once again, however, it is important to keep in
mind that each day, each location, and each target represent individual
events. There is an additional distinction made between releasing
arrested or detained individuals and releasing political prisoners
who have been convicted of political crimes. Releasing arrested
or detained individuals is a 501, whereas releasing political
prisoners convicted of political crimes is an event reflecting
a higher degree of cooperation or accommodation. As a result,
it is coded at a higher level on the cooperation scale; it is
a 601. In coding reports of surrenders, coders must be careful
to appropriately identify the actor. We can not assume that groups
are responsible for the actions of individual members. As a result,
in the example listed in section 2.6 which discusses the surrender
of six members of the M-19, we code the actor as unspecified guerrillas
(45) and the target as the state (10).
<P>
 5.7 Cessation of hostilities (502, 602, 703)
<P>
Events 502, 602, and 703 on the IPI cooperation scale all represent
the termination of violence, but with different degrees of scope
and permanence. Event 502, a cease fire, represents the cessation
of hostilities temporarily or in an isolated area. Event 602,
a truce, is an agreement between groups to end hostilities indefinitely.
The intention is to allow time for negotiations to see if the
conflict can be resolved. During this period, the armed struggle
is renounced. The signing of a peace treaty (event 703) implies
that the conflict has been permanently managed or resolved and
is often followed by disarmament.
<P>
5.8 Elections (504, 603, 704, 801, 902)
<P>
When a society has not been democratic and begins to institute
or reinstitute a democratic process into its political affairs,
the holding of elections represents a crucial cooperative event.
In societies in which democracy is institutionalized, elections
represent no more than the normal process of governing. Thus,
the IPI cooperation scale includes numerous codes for electoral
activity, but these are intended to apply only to societies without
institutionalized democratic systems, as anything constituting
the normal process of government is not codable under our system
(see section 2.1). The process of planning for and implementing
national elections involves many separate events. We code a concrete
agreement to allow elections to be held as a 504 on the cooperation
scale. When a date for national elections is set, or when minor
elections are held, we code a 603, &quot;implementing policy reform&quot;.

<P>
The election itself is codable as a 704, an 801, or a 902 depending
on the conditions. Any election held without full participation
(i.e., an election which any of the parties boycotts) is a 704.
The boycott would additionally be coded as a 404 on the conflict
scale with the party refusing to participate as actor and the
state as target. An election with full participation is coded
as either an 801 or a 902 depending on whether the election was
preceded by the development of a new constitution. In a society
where a democratic process is not institutionalized, the assumption
of power by the winners and the acceptance of the results by the
losers does not always follow elections without incident. As a
result, we recognize the act of allowing the opposition to take
power following an election as a separate cooperative event, coded
as a 603 on the cooperation scale with the state (10) as actor
and the general population (09) as target. Similarly, voiding
the results of an election is coded as an 807 on the conflict
scale, again with the general population (09) as target.  If an
election is not deemed free and fair by international observers,
we still consider the holding of the election to be a cooperative
event and code it as such. We additionally code one or more conflictual
events, however, to represent any violations of the electoral
process. 
<P>
6. USING THE CONFLICT AND COOPERATION SCALES TOGETHER
<P>
In some instances, the same action will represent cooperation
toward one target and conflict toward another. One common example
is a judicial decision, which generally represents cooperation
from the judicial branch toward the winner and conflict from the
judicial branch toward the loser. Consider this report from Latin
America Weekly Report, April 2, 1982: The labour courts judged
that the Cimetal management had not behaved responsibly and ruled
the strike legal, a highly unusual decision (page 7). In this
case, we code two events: one cooperative event (307) between
the judicial branch (13) and the labor union (55) and one conflictual
event (307) between the judicial branch (13) and business elites
(64). Another example in which both the conflict and cooperation
scales are used to code the same event was highlighted in section
2.3 above. When two or more groups coordinate their actions against
a common enemy, we code both conflictual events from the groups
to the common enemy, and cooperative events between the groups
themselves. It is important to note, however, that it is the coordination
of action which represents a cooperative event rather than the
action itself. Thus, if two groups protest against the government
together for three days, we code six conflictual events (one per
group per day), but only two cooperative events. We do not assume
a need to recoordinate policies each day, and thus we need not
repeat the coding of the cooperative events for additional days
of the same protest. The same assumption is invoked for events
which occur in several cities. If two groups protest together
in two cities, we code four conflictual events and two cooperative
events.
<P>
REFERENCES
<P>
Azar, Edward E. 1975. Ten Issues in Events Research. Theory and
Practice of Events Research. E. E. Azar and J. D. Ben-Dak, Eds.
New York: Gordon and Breech: 1-16.
<P>
Azar, Edward E. 1982. The Codebook of the Conflict and Peace Data
Bank. Center for International Development, University of Maryland.
<P>
Azar, Edward E. 1990. The Management of Protracted Social Conflict:
Theory and Cases. Brookfield, Vermont: USA Gower Publishing Company.

<P>
Burgess, P. M. and R. W. Lawton. 1972. Indicators of International
Behavior: An Assessment of Events Data Research. Beverly Hills:
Sage.
<P>
Davies, John L. 1993. The Global Event-Data System Manual. Center
for International Development and Conflict Management, University
of Maryland.
<P>
Gerner, Deborah J., Philip A. Schrodt, Ronald A. Francisco, and
Judith L. Weddle. 1992. Issues in the Creation of Event Data:
Machine Coding of Regional and International Sources. Paper presented
at the Annual Meeting of the American Political Science Association,
Chicago.
<P>
Goldstein, Joshua S. 1991. Reciprocity in Superpower Relations:
An Empirical Analysis. International Studies Quarterly 35: 195-209.
<P>
Gurr, Ted Robert. 1970. Why Men Rebel. Princeton: Princeton University
Press. 
<P>
Gurr, Ted Robert. 1993. Minorities at Risk: A Global View of Ethnopolitical
Conflicts. Washington, DC: United States Institute of Peace Press.

<P>
Howell, Llewellyn D. 1983. A Comparative Study of the WEIS and
COPDAB Data Sets. International Studies Quarterly 27: 149-159.
<P>
Keohane, Robert O. 1984. After Hegemony: Cooperation and Discord
in the World Political Economy. Princeton: Princeton University
Press.
<P>
McClelland, Charles. 1972. World Events Interaction Survey. Los
Angeles: University of Southern California, ICPR Edition.
<P>
Moore, Will H. and Ronny Lindstrom. 1994. The Violent Intranational
Conflict Data Project (VICDP) Codebook. Department of Political
Science, University of California- Riverside.
<P>
Muller, Edward N. and Mitchell A. Seligson. 1987. Inequality and
insurgency. American Political Science Review 81: 425-451.
<P>
Muller, Seligson, Fu, and Midlarsky. 1989. Land Equality and Political
Violence. American Political Science Review 83: 577-595.
<P>
Taylor, Charles Lewis and Michael C. Hudson. 1975. The World Handbook
of Political and Social Indicators, Second Edition. New Haven:
Yale University Press.
<P>
Taylor, Charles Lewis and David A. Jodice. 1983. World Handbook
of Political and Social Indicators. New Haven: Yale University
Press.
<P>
Taylor, Charles Lewis and David A. Jodice. 1985. The World Handbook
of Political and Social Indicators III, 1948-1982: Daily Political
Events Data. ICPSR Study #7761. Ann Arbor: Inter-University Consortium
for Political and Social Research. 
</BODY>

</HTML>
